Stripe Acquires OpenRouter and Targets Usage-Based AI Billing

Stripe agrees to acquire OpenRouter, the model aggregator claiming massive token volumes. The transaction, largely stock-based, aligns with a stated goal by Stripe: to manage the variability of inference costs and usage-based billing. OpenRouter's revenues are not disclosed, and the simulated order of magnitude illustrates the uncertainty surrounding the economic value of the traffic.

Undisclosed Price, Absent Revenues, and Primarily Stock Payment

The price mentioned for OpenRouter exceeds $8 billion, but Stripe's announcement provides no figures on revenue, margin, volume of credits sold, or the paid share of OpenRouter's traffic. The token volume does not allow for a direct revenue deduction due to free usage and the diversity of rates depending on the types of tokens and associated services. According to reports, most of the transaction's financing would be through Stripe stock. The company reached a valuation of $159 billion during a securities operation in February 2026, and the mentioned amount would correspond to about 5% of that valuation, subject to specific terms and references. Stripe is thus proceeding with the acquisition primarily using its own private shares, implying that the operation may not be fully liquid for OpenRouter's shareholders and resembles a share exchange.

Theoretical Orders of Magnitude Highlighting Uncertainty

Stripe claims that OpenRouter processes over 10 trillion tokens per day. For illustrative purposes, if all these tokens were charged at $1 per million, the annualized inference expense would reach approximately $3.65 billion, and a theoretical commission of 5.5% would be around $201 million. At $0.25 per million, this theoretical commission would be about $50 million, while at $5 it would exceed a billion. These amounts are not revenue estimates for OpenRouter: they ignore free models, negotiated terms, unused credits, discounts, client keys, and other billing methods. They primarily illustrate the uncertainty surrounding the economic value of traffic expressed in tokens.

Stated Goal: Managing Margin Beyond SaaS Revenue

Stripe indicates its intention to manage the variability of margins for AI services, shifting its intervention upstream of production costs with OpenRouter, even before the response generation. The company aims for the point where each request becomes a purchasing decision and sets its sights on aggregating the roles of purchasing center, meter, and billing system for intelligence. Historically, Stripe has already handled adjacent functions to revenue collection for SaaS, such as payments, subscriptions, declined card retries, invoicing, taxes, revenue recognition, and fraud prevention.

Why AI Disrupts Pricing and Margins for Publishers

In SaaS, the marginal cost per customer remains limited, and pricing is often per user. In contrast, in AI, each inference generates a cost, and pricing is expressed by usage, credit, task, or outcome, with a multitude of models and providers. The margin then depends on consumption and imposes upstream trade-offs. With each request, a model's response requires a new calculation and solicits resources such as tokens, computing power, memory, and sometimes third-party services. The cost depends on several factors: demand, the model used, context length, the number of reasoning steps, and the number of attempts. Thus, two applications with identical revenue can present very different margins; a subscription at €20 can remain profitable with rare usage and become unprofitable with daily use of an expensive model and long contexts.

OpenRouter: From Multi-Model Gateway to Fintech Mechanism

OpenRouter is an interface that allows access to multiple models without having to separately integrate APIs from players like OpenAI, Anthropic, Google, or DeepSeek. The platform operates with dollar credits: the user credits an account, and the cost of each request is deducted from the balance. OpenRouter claims to pass on the public rates of providers without markup on inference; however, it charges 5.5% when purchasing credits, with a minimum of $0.80, and may bill clients using their own keys beyond certain thresholds. A router can also select a model based on the requested complexity, price, speed, availability, or company privacy rules.

An Installed Base and Usage Data at the Core of the Asset

Stripe announced the acquisition of OpenRouter, which provides access to over 400 models from more than 80 providers and claims over 10 million developers and businesses. In January 2026, OpenRouter was already using Stripe for billing, taxes, and fraud prevention, boasting five million developers at that time. The asset includes a broad distribution among developers, a position between model providers, and economic data detailing the models used, accepted prices, provider changes, latency, failure rates, and sensitivity to cost/quality trade-offs. While it is possible to build a router, moving millions of existing integrations is much more challenging.